Search

We consider the problem of automatically decomposing operations over tensors or arrays so that they can be executed in parallel on multiple devices. We address two, closely-linked questions. First, what programming abstraction should systems for tensor-based computing offer to enable such decompositions? Second, given that abstraction, how should such systems automatically decompose a tensor-based computation? We assert that tensor-based systems should offer a programming abstraction based on an extended Einstein summation notation, which is a fully declarative, mathematical specification for tensor computations. We show that any computation specified in the Einstein summation notation can be re-written into an equivalent tensor-relational computation, and this re-write generalizes existing notations of tensor parallelism such as "data parallel'' and "model parallel.'' We consider the algorithmic problem of optimally computing a tensor-relational decomposition of a graph of operations specified in our extended Einstein summation notation, and we experimentally show the value of the algorithm that we develop.

Large language models (LLMs) have recently emerged as promising tools for solving challenging robotic tasks, even in the presence of action and observation uncertainties. Recent LLM-based decision-making methods (also referred to as LLM-based agents), when paired with appropriate critics, have demonstrated potential in solving complex, long-horizon tasks with relatively few interactions. However, most existing LLM-based agents lack the ability to retain and learn from past interactions - an essential trait of learning-based robotic systems. We propose RAG-Modulo, a framework that enhances LLM-based agents with a memory of past interactions and incorporates critics to evaluate the agents' decisions. The memory component allows the agent to automatically retrieve and incorporate relevant past experiences as in-context examples, providing context-aware feedback for more informed decision-making. Further by updating its memory, the agent improves its performance over time, thereby exhibiting learning. Through experiments in the challenging BabyAI and AlfWorld domains, we demonstrate significant improvements in task success rates and efficiency, showing that the proposed RAG-Modulo framework outperforms state-of-the-art baselines., Comment: 8 pages, 5 figures

Parameter-Efficient Fine-Tuning (PEFT) has become the standard for customising Foundation Models (FMs) to user-specific downstream tasks. However, typical PEFT methods require storing multiple task-specific adapters, creating scalability issues as these adapters must be housed and run at the FM server. Traditional prompt tuning offers a potential solution by customising them through task-specific input prefixes, but it under-performs compared to other PEFT methods like LoRA. To address this gap, we propose Low-Rank Prompt Adaptation (LoPA), a prompt-tuning-based approach that performs on par with state-of-the-art PEFT methods and full fine-tuning while being more parameter-efficient and not requiring a server-based adapter. LoPA generates soft prompts by balancing between sharing task-specific information across instances and customization for each instance. It uses a low-rank decomposition of the soft-prompt component encoded for each instance to achieve parameter efficiency. We provide a comprehensive evaluation on multiple natural language understanding and code generation and understanding tasks across a wide range of foundation models with varying sizes., Comment: 14 pages, 8 figures, 4 tables

We carefully evaluate a number of algorithms for learning in a federated environment, and test their utility for a variety of image classification tasks. We consider many issues that have not been adequately considered before: whether learning over data sets that do not have diverse sets of images affects the results; whether to use a pre-trained feature extraction "backbone"; how to evaluate learner performance (we argue that classification accuracy is not enough), among others. Overall, across a wide variety of settings, we find that vertically decomposing a neural network seems to give the best results, and outperforms more standard reconciliation-used methods., Comment: 16 pages, 7 figures, Accepted at ICCV2023

The relational data model was designed to facilitate large-scale data management and analytics. We consider the problem of how to differentiate computations expressed relationally. We show experimentally that a relational engine running an auto-differentiated relational algorithm can easily scale to very large datasets, and is competitive with state-of-the-art, special-purpose systems for large-scale distributed machine learning., Comment: ICML 2023

Large Language Models (LLMs) pre-trained on code have recently emerged as the dominant approach to program synthesis. However, these models are trained using next-token prediction, which ignores the syntax and semantics of code. We propose RLCF, that further trains a pre-trained LLM via reinforcement learning, using feedback from a grounding function that scores the quality of the code. The grounding function uses (i) compiler-derived feedback on whether the code it generates passes a set of correctness checks; and (ii) feedback from a different LLM that compares the generated code to a reference code. RLCF is model- and language-agnostic. We empirically evaluate it on the MBJP and MathQA tasks for Java. Our experiments show that RLCF raises the odds that an LLM-generated program compiles, is executable, and produces the right output on tests, often allowing LLMs to match the performance of 2x-8x larger LLMs., Comment: 23 pages

Recent work on the Lottery Ticket Hypothesis (LTH) shows that there exist ``\textit{winning tickets}'' in large neural networks. These tickets represent ``sparse'' versions of the full model that can be trained independently to achieve comparable accuracy with respect to the full model. However, finding the winning tickets requires one to \emph{pretrain} the large model for at least a number of epochs, which can be a burdensome task, especially when the original neural network gets larger. In this paper, we explore how one can efficiently identify the emergence of such winning tickets, and use this observation to design efficient pretraining algorithms. For clarity of exposition, our focus is on convolutional neural networks (CNNs). To identify good filters, we propose a novel filter distance metric that well-represents the model convergence. As our theory dictates, our filter analysis behaves consistently with recent findings of neural network learning dynamics. Motivated by these observations, we present the \emph{LOttery ticket through Filter-wise Training} algorithm, dubbed as \textsc{LoFT}. \textsc{LoFT} is a model-parallel pretraining algorithm that partitions convolutional layers by filters to train them independently in a distributed setting, resulting in reduced memory and communication costs during pretraining. Experiments show that \textsc{LoFT} $i)$ preserves and finds good lottery tickets, while $ii)$ it achieves non-trivial computation and communication savings, and maintains comparable or even better accuracy than other pretraining methods.

Asynchronous learning protocols have regained attention lately, especially in the Federated Learning (FL) setup, where slower clients can severely impede the learning process. Herein, we propose \texttt{AsyncDrop}, a novel asynchronous FL framework that utilizes dropout regularization to handle device heterogeneity in distributed settings. Overall, \texttt{AsyncDrop} achieves better performance compared to state of the art asynchronous methodologies, while resulting in less communication and training time overheads. The key idea revolves around creating ``submodels'' out of the global model, and distributing their training to workers, based on device heterogeneity. We rigorously justify that such an approach can be theoretically characterized. We implement our approach and compare it against other asynchronous baselines, both by design and by adapting existing synchronous FL algorithms to asynchronous scenarios. Empirically, \texttt{AsyncDrop} reduces the communication cost and training time, while matching or improving the final test accuracy in diverse non-i.i.d. FL scenarios.

State-of-the-art neural models of source code tend to be evaluated on the generation of individual expressions and lines of code, and commonly fail on long-horizon tasks such as the generation of entire method bodies. We propose to address this deficiency using weak supervision from a static program analyzer. Our neurosymbolic method allows a deep generative model to symbolically compute, using calls to a static-analysis tool, long-distance semantic relationships in the code that it has already generated. During training, the model observes these relationships and learns to generate programs conditioned on them. We apply our approach to the problem of generating entire Java methods given the remainder of the class that contains the method. Our experiments show that the approach substantially outperforms state-of-the-art transformers and a model that explicitly tries to learn program semantics on this task, both in terms of producing programs free of basic semantic errors and in terms of syntactically matching the ground truth., Comment: Accepted for publication at Neurips 2021

Recent papers have suggested that transfer learning can outperform sophisticated meta-learning methods for few-shot image classification. We take this hypothesis to its logical conclusion, and suggest the use of an ensemble of high-quality, pre-trained feature extractors for few-shot image classification. We show experimentally that a library of pre-trained feature extractors combined with a simple feed-forward network learned with an L2-regularizer can be an excellent option for solving cross-domain few-shot image classification. Our experimental results suggest that this simpler sample-efficient approach far outperforms several well-established meta-learning algorithms on a variety of few-shot tasks., Comment: 18 pages including appendix and references. 2 figures in the main paper, 1 figure in appendix. Accepted at ICCV 2021

We consider the question: what is the abstraction that should be implemented by the computational engine of a machine learning system? Current machine learning systems typically push whole tensors through a series of compute kernels such as matrix multiplications or activation functions, where each kernel runs on an AI accelerator (ASIC) such as a GPU. This implementation abstraction provides little built-in support for ML systems to scale past a single machine, or for handling large models with matrices or tensors that do not easily fit into the RAM of an ASIC. In this paper, we present an alternative implementation abstraction called the tensor relational algebra (TRA). The TRA is a set-based algebra based on the relational algebra. Expressions in the TRA operate over binary tensor relations, where keys are multi-dimensional arrays and values are tensors. The TRA is easily executed with high efficiency in a parallel or distributed environment, and amenable to automatic optimization. Our empirical study shows that the optimized TRA-based back-end can significantly outperform alternatives for running ML workflows in distributed clusters.

Persistent partitioning is effective in avoiding expensive shuffling operations. However it remains a significant challenge to automate this process for Big Data analytics workloads that extensively use user defined functions (UDFs), where sub-computations are hard to be reused for partitionings compared to relational applications. In addition, functional dependency that is widely utilized for partitioning selection is often unavailable in the unstructured data that is ubiquitous in UDF-centric analytics. We propose the Lachesis system, which represents UDF-centric workloads as workflows of analyzable and reusable sub-computations. Lachesis further adopts a deep reinforcement learning model to infer which sub-computations should be used to partition the underlying data. This analysis is then applied to automatically optimize the storage of the data across applications to improve the performance and users' productivity., Comment: In submission

We present a new approach, called meta-meta classification, to learning in small-data settings. In this approach, one uses a large set of learning problems to design an ensemble of learners, where each learner has high bias and low variance and is skilled at solving a specific type of learning problem. The meta-meta classifier learns how to examine a given learning problem and combine the various learners to solve the problem. The meta-meta learning approach is especially suited to solving few-shot learning tasks, as it is easier to learn to classify a new learning problem with little data than it is to apply a learning algorithm to a small data set. We evaluate the approach on a one-shot, one-class-versus-all classification task and show that it is able to outperform traditional meta-learning as well as ensembling approaches., Comment: 10 pages without references, 3 figures

Consider the case where a programmer has written some part of a program, but has left part of the program (such as a method or a function body) incomplete. The goal is to use the context surrounding the missing code to automatically 'figure out' which of the codes in the database would be useful to the programmer in order to help complete the missing code. The search is 'contextualized' in the sense that the search engine should use clues in the partially-completed code to figure out which database code is most useful. The user should not be required to formulate an explicit query. We cast contextualized code search as a learning problem, where the goal is to learn a distribution function computing the likelihood that each database code completes the program, and propose a neural model for predicting which database code is likely to be most useful. Because it will be prohibitively expensive to apply a neural model to each code in a database of millions or billions of codes at search time, one of our key technical concerns is ensuring a speedy search. We address this by learning a 'reverse encoder' that can be used to reduce the problem of evaluating each database code to computing a convolution of two normal distributions.

A number of popular systems, most notably Google's TensorFlow, have been implemented from the ground up to support machine learning tasks. We consider how to make a very small set of changes to a modern relational database management system (RDBMS) to make it suitable for distributed learning computations. Changes include adding better support for recursion, and optimization and execution of very large compute plans. We also show that there are key advantages to using an RDBMS as a machine learning platform. In particular, learning based on a database management system allows for trivial scaling to large data sets and especially large models, where different computational units operate on different parts of a model that may be too large to fit into RAM.

Storage and memory systems for modern data analytics are heavily layered, managing shared persistent data, cached data, and non-shared execution data in separate systems such as distributed file system like HDFS, in-memory file system like Alluxio and computation framework like Spark. Such layering introduces significant performance and management costs for copying data across layers redundantly and deciding proper resource allocation for all layers. In this paper we propose a single system called Pangea that can manage all data---both intermediate and long-lived data, and their buffer/caching, data placement optimization, and failure recovery---all in one monolithic storage system, without any layering. We present a detailed performance evaluation of Pangea and show that its performance compares favorably with several widely used layered systems such as Spark., Comment: 14 pages

This paper describes PlinyCompute, a system for development of high-performance, data-intensive, distributed computing tools and libraries. In the large, PlinyCompute presents the programmer with a very high-level, declarative interface, relying on automatic, relational-database style optimization to figure out how to stage distributed computations. However, in the small, PlinyCompute presents the capable systems programmer with a persistent object data model and API (the "PC object model") and associated memory management system that has been designed from the ground-up for high performance, distributed, data-intensive computing. This contrasts with most other Big Data systems, which are constructed on top of the Java Virtual Machine (JVM), and hence must at least partially cede performance-critical concerns such as memory management (including layout and de/allocation) and virtual method/function dispatch to the JVM. This hybrid approach---declarative in the large, trusting the programmer's ability to utilize PC object model efficiently in the small---results in a system that is ideal for the development of reusable, data-intensive tools and libraries. Through extensive benchmarking, we show that implementing complex objects manipulation and non-trivial, library-style computations on top of PlinyCompute can result in a speedup of 2x to more than 50x or more compared to equivalent implementations on Spark., Comment: 48 pages, including references and Appendix

We introduce program splicing, a programming methodology that aims to automate the commonly used workflow of copying, pasting, and modifying code available online. Here, the programmer starts by writing a "draft" that mixes unfinished code, natural language comments, and correctness requirements in the form of test cases or API call sequence constraints. A program synthesizer that interacts with a large, searchable database of program snippets is used to automatically complete the draft into a program that meets the requirements. The synthesis process happens in two stages. First, the synthesizer identifies a small number of programs in the database that are relevant to the synthesis task. Next it uses an enumerative search to systematically fill the draft with expressions and statements from these relevant programs. The resulting program is returned to the programmer, who can modify it and possibly invoke additional rounds of synthesis. We present an implementation of program splicing for the Java programming language. The implementation uses a corpus of over 3.5 million procedures from an open-source software repository. Our evaluation uses the system in a suite of everyday programming tasks, and includes a comparison with a state-of-the-art competing approach as well as a user study. The results point to the broad scope and scalability of program splicing and indicate that the approach can significantly boost programmer productivity.

We study the problem of generating source code in a strongly typed, Java-like programming language, given a label (for example a set of API calls or types) carrying a small amount of information about the code that is desired. The generated programs are expected to respect a "realistic" relationship between programs and labels, as exemplified by a corpus of labeled programs available during training. Two challenges in such conditional program generation are that the generated programs must satisfy a rich set of syntactic and semantic constraints, and that source code contains many low-level features that impede learning. We address these problems by training a neural generator not on code but on program sketches, or models of program syntax that abstract out names and operations that do not generalize across programs. During generation, we infer a posterior distribution over sketches, then concretize samples from this distribution into type-safe programs using combinatorial techniques. We implement our ideas in a system for generating API-heavy Java code, and show that it can often predict the entire body of a method given just a few API calls or data types that appear in the method.

We present a Bayesian framework for learning probabilistic specifications from large, unstructured code corpora, and a method to use this framework to statically detect anomalous, hence likely buggy, program behavior. The distinctive insight here is to build a statistical model that correlates all specifications hidden inside a corpus with the syntax and observed behavior of programs that implement these specifications. During the analysis of a particular program, this model is conditioned into a posterior distribution that prioritizes specifications that are relevant to this program. This allows accurate program analysis even if the corpus is highly heterogeneous. The problem of finding anomalies is now framed quantitatively, as a problem of computing a distance between a "reference distribution" over program behaviors that our model expects from the program, and the distribution over behaviors that the program actually produces. We present a concrete embodiment of our framework that combines a topic model and a neural network model to learn specifications, and queries the learned models to compute anomaly scores. We evaluate this implementation on the task of detecting anomalous usage of Android APIs. Our encouraging experimental results show that the method can automatically discover subtle errors in Android applications in the wild, and has high precision and recall compared to competing probabilistic approaches.

As of 2005, sampling has been incorporated in all major database systems. While efficient sampling techniques are realizable, determining the accuracy of an estimate obtained from the sample is still an unresolved problem. In this paper, we present a theoretical framework that allows an elegant treatment of the problem. We base our work on generalized uniform sampling (GUS), a class of sampling methods that subsumes a wide variety of sampling techniques. We introduce a key notion of equivalence that allows GUS sampling operators to commute with selection and join, and derivation of confidence intervals. We illustrate the theory through extensive examples and give indications on how to use it to provide meaningful estimations in database systems.